Strange KEEP behavior in linker script

[Denis FERRANTI]

Well, my issue is the fact that the output section is discarded.
I tried the --no-gc-sections but it has no effect.
I can stop the linker from discarding the section adding a contribution 
like below, but that's ....strange ;-)

.mysec :
{
   . += 1;                  /* dummy contribution */
    mysec_start = .;

    KEEP(*(.mysec));

    mysec_end = .;
}

When there are no contributions, the section seems to be removed in 
ldlang.c:

strip_excluded_output_sections (void)
{
...
      exclude = (output_section->rawsize == 0
         && (output_section->flags & SEC_KEEP) == 0
         && !bfd_section_removed_from_list (output_bfd,
                            output_section));
...
      if (exclude)
    {
      /* We don't set bfd_section to NULL since bfd_section of the
         removed output section statement may still be used.  */
      if (!os->section_relative_symbol
          && !os->update_dot_tree)
        os->ignored = TRUE;
            output_section->flags |= SEC_EXCLUDE;
      bfd_section_list_remove (output_bfd, output_section);
      output_bfd->section_count--;
    }
...
}
The SEC_KEEP flag is tested on the output section.

I see your point about the KEEP() semantic, but then do you see a 
solution to avoid the discard of the output section (without adding 
artificial contributions in the linker script)?


Daniel Jacobowitz wrote:
> On Tue, May 06, 2008 at 03:29:27PM +0200, Denis FERRANTI wrote:
>   
>> Shouldn't the KEEP() statement force the generation of the output section 
>> in the ELF file, even if there are no input sections.
>>     
>
> No, why should it?  It is supposed to keep the input sections, not the
> output section.  I'm not sure why ld started discarding the empty
> output section.
>
>   

-- 
Denis Ferranti